Lesion Segmentation Method Based on Deep Learning CT Image of Pulmonary Tuberculosis

Tuberculosis is a major public health problem that is the leading cause of death worldwide. Early detection and diagnosis is the key to the treatment of tuberculosis. Computed tomography (CT) can provide more comprehensive tuberculosis lesion information and improve the accuracy of diagnosis. However, due to the characteristics of polymorphism, multiple parts, multiple nodules and cavities of pulmonary tuberculosis, segmentation has become an important and difficult problem in computer-aided diagnosis.Deep learning is widely used in medical image segmentation tasks. This paper proposes to use U-Net and attention mechanism to form Attention U-Net network model for feature extraction and segmentation of labeled CT images of tuberculosis, to achieve unlabeled tuberculosis CT image data Perform lesion segmentation and lesion labeling.